vyre-libs 0.6.2

vyre Category A library ecosystem - pure-IR compositions over vyre-ops hardware primitives
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206

//! Tiled linear-layer constructors (`linear_tiled`, `linear_tiled_reference`).

use vyre::ir::{BufferAccess, BufferDecl, DataType, Expr, Node, Program};

use crate::region::{tag_program, wrap};
use crate::tensor_ref::TensorRef;
use crate::MatmulBiasTiled;

use super::builder::linear;

pub(super) const LINEAR_TILED_OP_ID: &str = "vyre-libs::nn::linear_tiled";
pub(super) const LINEAR_TILED_REFERENCE_OP_ID: &str = "vyre-libs::nn::linear_tiled_reference";
pub(super) const LINEAR_TILED_TILE: u32 = 32;
pub(super) const LINEAR_TILED_MIN_WORK: u32 = 1024;

/// Build a tiled linear-layer Program: `out[j] = b[j] + sum_k x[k] * w[k, j]`.
///
/// # Errors
/// Returns `Err` when dimensions are empty, overflow buffer counts, or `tile == 0`.
pub fn linear_tiled(
    x: &str,
    w: &str,
    b: &str,
    out: &str,
    in_dim: u32,
    out_dim: u32,
    tile: u32,
) -> Result<Program, String> {
    if in_dim == 0 {
        return Err("Fix: linear_tiled in_dim=0 is invalid: empty reduction".to_string());
    }
    if out_dim == 0 {
        return Err("Fix: linear_tiled out_dim=0 is invalid: empty output".to_string());
    }
    if tile == 0 {
        return Err("Fix: linear_tiled tile=0 is invalid: tile width must be > 0".to_string());
    }
    in_dim.checked_mul(out_dim).ok_or_else(|| {
        "Fix: linear_tiled in_dim*out_dim overflows u32; reduce dimensions.".to_string()
    })?;
    let program = MatmulBiasTiled::new(
        TensorRef::u32_2d(x, 1, in_dim),
        TensorRef::u32_2d(w, in_dim, out_dim),
        TensorRef::u32_1d(b, out_dim),
        TensorRef::u32_2d(out, 1, out_dim),
        tile,
    )
    .build()
    .map_err(|error| format!("Fix: linear_tiled matmul_tiled build failed: {error}"))?;
    Ok(tag_program(LINEAR_TILED_OP_ID, program))
}

/// Reference / oracle implementation of tiled linear (hand-rolled IR).
/// Kept for parity testing against the optimized `linear_tiled` path.
#[allow(clippy::too_many_arguments)]
pub fn linear_tiled_reference(
    x: &str,
    w: &str,
    b: &str,
    out: &str,
    in_dim: u32,
    out_dim: u32,
    tile: u32,
) -> Result<Program, String> {
    if in_dim == 0 {
        return Err("Fix: linear_tiled_reference in_dim=0 is invalid: empty reduction".to_string());
    }
    if out_dim == 0 {
        return Err("Fix: linear_tiled_reference out_dim=0 is invalid: empty output".to_string());
    }
    if tile == 0 {
        return Err(
            "Fix: linear_tiled_reference tile=0 is invalid: tile width must be > 0".to_string(),
        );
    }
    let weight_count = in_dim.checked_mul(out_dim).ok_or_else(|| {
        "Fix: linear_tiled_reference in_dim*out_dim overflows u32; reduce dimensions.".to_string()
    })?;
    let tile_count = in_dim.div_ceil(tile);
    let lane = Expr::var("lane");
    let kk = Expr::var("kk");
    let body = vec![
        Node::let_bind("lane", Expr::InvocationId { axis: 0 }),
        Node::if_then(
            Expr::lt(lane.clone(), Expr::u32(out_dim)),
            vec![
                Node::let_bind("acc", Expr::load(b, lane.clone())),
                Node::loop_for(
                    "tile_idx",
                    Expr::u32(0),
                    Expr::u32(tile_count),
                    vec![
                        Node::let_bind(
                            "tile_base",
                            Expr::mul(Expr::var("tile_idx"), Expr::u32(tile)),
                        ),
                        Node::loop_for(
                            "tile_k",
                            Expr::u32(0),
                            Expr::u32(tile),
                            vec![
                                Node::let_bind(
                                    "kk",
                                    Expr::add(Expr::var("tile_base"), Expr::var("tile_k")),
                                ),
                                Node::if_then(
                                    Expr::lt(kk.clone(), Expr::u32(in_dim)),
                                    vec![Node::assign(
                                        "acc",
                                        Expr::add(
                                            Expr::var("acc"),
                                            Expr::mul(
                                                Expr::load(x, kk.clone()),
                                                Expr::load(
                                                    w,
                                                    Expr::add(
                                                        Expr::mul(kk.clone(), Expr::u32(out_dim)),
                                                        lane.clone(),
                                                    ),
                                                ),
                                            ),
                                        ),
                                    )],
                                ),
                            ],
                        ),
                    ],
                ),
                Node::Store {
                    buffer: out.into(),
                    index: lane,
                    value: Expr::var("acc"),
                },
            ],
        ),
    ];
    Ok(Program::wrapped(
        vec![
            BufferDecl::storage(x, 0, BufferAccess::ReadOnly, DataType::U32).with_count(in_dim),
            BufferDecl::storage(w, 1, BufferAccess::ReadOnly, DataType::U32)
                .with_count(weight_count),
            BufferDecl::storage(b, 2, BufferAccess::ReadOnly, DataType::U32).with_count(out_dim),
            BufferDecl::output(out, 3, DataType::U32).with_count(out_dim),
        ],
        [256, 1, 1],
        vec![wrap(LINEAR_TILED_REFERENCE_OP_ID, body, None)],
    ))
}

inventory::submit! {
    crate::harness::OpEntry {
        id: "vyre-libs::nn::linear",
        build: || {
            linear("x", "w", "b", "out", 4, 4)
                .unwrap_or_else(|error| crate::invalid_program("vyre-libs::nn::linear", format!("Fix: linear fixture dimensions are invalid: {error}")))
        },
        // V7-TEST-005: deterministic fixture for linear(4, 4).
        // Body indexes `w[k * out_dim + i]` (column-major per out_dim),
        // so for w = [0..16], out_dim = 4:
        //   out[i] = b[i] + sum_k x[k] * w[k*4 + i]
        // With x = [0, 1, 2, 3] and b = [0, 0, 0, 0]:
        //   out[0] = 0*0 + 1*4 + 2*8  + 3*12 =  4 + 16 + 36 = 56
        //   out[1] = 0*1 + 1*5 + 2*9  + 3*13 =  5 + 18 + 39 = 62
        //   out[2] = 0*2 + 1*6 + 2*10 + 3*14 =  6 + 20 + 42 = 68
        //   out[3] = 0*3 + 1*7 + 2*11 + 3*15 =  7 + 22 + 45 = 74
        test_inputs: Some(|| {

            let x = crate::test_support::byte_pack::u32_bytes(&(0..4).collect::<Vec<_>>());
            let w = crate::test_support::byte_pack::u32_bytes(&(0..16).collect::<Vec<_>>());
            let bias = crate::test_support::byte_pack::u32_bytes(&[0, 0, 0, 0]);
            // The output buffer is declared with `with_count(out_dim) = 4`
            // u32s = 16 bytes. The CPU reference and the GPU dispatch both
            // honor that buffer length; an over-allocated input slot would
            // make CPU echo a longer Value than the GPU returns and trip
            // the CPU/GPU length divergence assertion in cat_a_gpu_differential.
            vec![vec![x, w, bias]]
        }),
        expected_output: Some(|| {

            vec![vec![crate::test_support::byte_pack::u32_bytes(&[56, 62, 68, 74])]]
        }),
        category: Some("nn"),
    }
}

inventory::submit! {
    crate::harness::OpEntry {
        id: "vyre-libs::nn::linear_tiled",
        build: || {
            linear_tiled("x", "w", "b", "out", 4, 4, 2)
                .unwrap_or_else(|error| crate::invalid_program("vyre-libs::nn::linear_tiled", format!("Fix: linear_tiled fixture dimensions are invalid: {error}")))
        },
        test_inputs: Some(|| {

            let x = crate::test_support::byte_pack::u32_bytes(&(0..4).collect::<Vec<_>>());
            let w = crate::test_support::byte_pack::u32_bytes(&(0..16).collect::<Vec<_>>());
            let bias = crate::test_support::byte_pack::u32_bytes(&[0, 0, 0, 0]);
            vec![vec![x, w, bias]]
        }),
        expected_output: Some(|| {

            vec![vec![crate::test_support::byte_pack::u32_bytes(&[56, 62, 68, 74])]]
        }),
        category: Some("nn"),
    }
}